in hops.util.findfringe
currently frequencies are assumed to stack into a regular grid
current DiFX configuration for EHT 2017 breaks this by having discontiguous HOPS channels, with subchannels that do not lie on a regular grid
an FFT fringe search over the entire band needs to be adapted to (approximately) handle this
one option is to place visibilities onto a supersampled regular grid, which is memory intensive but will scale computationally as logN which isn't so bad
can possibly deal with the memory issue by making sure we loop efficiently over time, at each stage we can throw away the "zero-padded" part of the delay spectrum

update fringe solution globalization to include known R-L delays, delay priors
update bin/closecf or begin a new set of routines

in hops.util.findfringe
currently fourier interpolation is used to zoom into the fringe location but this is inefficient to apply to the entire search space
add additional stage quadratic interpolation
or put fringe at DC and do a limited supersampling, or do a direct search
can plot statistical uncertainties as well

bin/closecf returns a maximum-likelihood solution
adapt this to generate a fourfit search window, including statistical and fixed systematic uncertainties

For EHT 2015+ data, DiFX is changed to use 1 site label for dual-polarization stations, instead of separate site labels.
closecf should be updated for this new book-keeping

The calculation of effective averaging timescale to use for atmospheric phase calibration,

is based on the Eq 21 from https://arxiv.org/abs/1903.08832

The original formula should be double checked.